Quantitative analysis method for steam turbine set based on dimension reduction parameter reconstruction of hash function

Abstract

The application discloses a quantitative analysis method for steam turbine unit based on dimension reduction parameter reconstruction of a hash function, relates to the technical field of power machinery fault diagnosis, and solves the technical problems that dimension reduction is not thorough enough when high-dimensional data space of a steam turbine unit is reconstructed in parameters, and a quantitative function relation between a health factor and a health state of the steam turbine unit is not constructed.The technical scheme points are that a relation between dynamic change characteristics of working condition fingerprint time series information entropy and a system health factor is analyzed in depth, a health factor degradation law of a whole life cycle of the steam turbine unit is revealed, a working condition fingerprint complexity entropy increase formation mechanism of the steam turbine unit is revealed, a health factor quantitative evaluation model of the steam turbine unit is established, health factor quantitative calculation of the steam turbine unit is realized, and the application problem of PHM technology in engineering equipment such as the steam turbine unit is solved, so that scientific basis and technical support are provided for research work in the field of quantitative maintenance and health management of the steam turbine unit.

Description

A Quantitative Analysis Method for Steam Turbine Units Based on Hash Function Dimensionality Reduction Parameter Reconstruction Technical Field

[0001] This application relates to the field of power machinery fault diagnosis technology, and in particular to a quantitative analysis method for steam turbine sets based on hash function-based dimensionality reduction parameter reconstruction. Background Technology

[0002] Fault diagnosis and health management (PHM) technology aims to replace reactive maintenance with proactive maintenance, which can effectively improve the operational reliability and service life of steam turbine units.

[0003] Current research on steam turbine units largely focuses on fault diagnosis, while research on health management is still in the exploratory stage. The main challenges lie in the lack of a unified quantitative evaluation standard for health management technology, and the unclear definition of the characteristic parameter "health factor." These issues directly limit the application of PHM technology in engineering equipment such as steam turbine units. Current international research progress in fault diagnosis and health management of complex systems such as steam turbine units mainly focuses on the following aspects:

[0004] (1) Knowledge-based methods: Knowledge-based methods mainly rely on the basic judgment experience of operation and maintenance personnel on the operating status of equipment in long-term work. Commonly used methods include expert systems, hierarchical analysis, and fuzzy evaluation methods. Knowledge-based methods are suitable for studying certain operating characteristics or working conditions and are widely used in the fault diagnosis and fault prediction of steam turbines. However, they are difficult to use for the study of "health factors" of equipment with unclear specific working conditions.

[0005] (2) State model-based method: The state model-based method is a physical or mathematical model with determinism and specificity established on the basis of a full understanding of the operation and degradation mechanism of the equipment. The state model-based method is suitable for the analysis of simplified and abstract dynamic models and is widely used in the study of dynamic characteristics and failure mechanism of steam turbine units. However, the quantitative health management of equipment requires the combination of a large number of measurement points of field sensors, which leads to the need for repeated model correction based on individual differences, and its generalization ability is not ideal.

[0006] (3) Data-driven approach: The data-driven approach mainly utilizes the monitoring data of various state parameters collected by sensors to drive various classification methods or data mining methods to analyze and evaluate the health status of equipment. Research results in this field are abundant, mainly including: grey system theory, neural networks, support vector machines, hidden Markov models, time series models, autoregressive models, logistic regression models, kernel function methods, etc. The data-driven approach does not require prior knowledge of the target system (state models and experience, etc.). Based on the data collected by sensors, it mines the implicit information through various data analysis and processing methods, avoiding the shortcomings of the above two methods. It is a relatively ideal method for fault prediction and health management. However, in specific application areas, the insufficient fault data samples restrict the development of this method in the field of turbine unit fault prediction.

[0007] In summary, current kernel transformation methods can transform high-dimensional spatial data into low-dimensional data through mapping, but the dimensionality reduction is not thorough enough for the high-dimensional data space of steam turbine units during parameter reconstruction. Furthermore, a health factor (HI) consistent with the health and degradation states of the steam turbine unit system has not yet been established, and a quantitative functional relationship between HI and the health state of the steam turbine unit is lacking. Therefore, a quantitative analysis method for steam turbine units based on hash function-based dimensionality reduction and parameter reconstruction is proposed. Summary of the Invention

[0008] This application provides a quantitative analysis method for steam turbine units based on hash function-based dimensionality reduction parameter reconstruction. The technical objective is to utilize the dynamic value sequence of health factors as a data source, conduct online experiments, and carry out relevant research on the effectiveness of quantitative calculation methods for health factors and system performance degradation models. This aims to improve the calculation method, enhance accuracy, and ultimately solve the quantitative evaluation problem of health management technology in the PHM field. It also provides a new approach for the transformation of health management technology for complex systems from qualitative signal analysis to quantitative informatics analysis methods.

[0009] The above-mentioned technical objective of this application is achieved through the following technical solution:

[0010] A quantitative analysis method for steam turbine units based on hash function-based dimensionality reduction parameter reconstruction includes:

[0011] S1: Analyze the types and number of sensors in the experimental setup used for quantitative analysis of steam turbine units to obtain sensor parameters;

[0012] S2: Construct a multidimensional variable storage matrix using historical operational data and sensor parameters;

[0013] S3: An improved hash function algorithm is used to perform collision and perturbation sensitivity analysis on the multidimensional variable storage matrix to obtain one-dimensional working condition fingerprint sequence data;

[0014] S4: Perform dimensionality reduction projection completeness analysis on the one-dimensional working condition fingerprint sequence data to generate the working condition fingerprint sequence information entropy;

[0015] S5: Perform quantitative analysis of health factors based on the entropy of the working condition fingerprint sequence information to obtain quantitative values ​​of health factors;

[0016] S6: Analyze the correlation between quantitative values ​​of health factors and health status based on data-driven methods to construct a quantitative health factor model for steam turbine units;

[0017] S7: Conduct online verification experiments on the quantitative health factor model of the turbine set using historical operating data to optimize and improve the quantitative health factor model of the turbine set;

[0018] S8: Quantitative analysis of the turbine set is conducted using the optimized and improved quantitative health factor model.

[0019] Furthermore, in step S3, the improved hash function algorithm includes:

[0020] The sampled values ​​of each sensor stored in the multidimensional variable storage matrix are normalized to obtain normalized parameters;

[0021] The number of significant digits i in the normalization parameter is analyzed, and then the normalization parameter is multiplied by 10i to convert it into a large integer value, which is the effective input value of the improved hash function algorithm.

[0022] In this system, the sampled values ​​from each sensor are binary digital signals converted from analog to digital signals (A / D converters). If the analog input of the A / D converter is x, the digital output is y, and the ratio of the transmitter's input range to its output range is K, then... m Then y = x / K m q; where q represents the quantization unit of the A / D conversion, and q = M / 2 N M represents the range of the analog input quantity of the A / D conversion, and N represents the number of bits in the A / D conversion.

[0023] Furthermore, in step S3, the one-dimensional working condition fingerprint sequence data is obtained by mapping the array D[m] to an integer space represented by 8×n bits.

[0024] Further, in step S4, the entropy of the working condition fingerprint sequence information is represented as:

[0025] H(X)=-∑ x∈X P(x)log2P(x);(1)

[0026] Where H(X) represents the information entropy value, and P(x)log2P(x) represents the uncertainty of the system state.

[0027] Furthermore, in step S5, the quantitative value of the health factor is a dynamically weighted analysis of the statistical probability in equation (1), expressed as:

[0028] H W (X)=-∑ x∈X W(P)P(x)log2[W(P)P(x)];

[0029] Where W(P) represents the probability statistical distribution function of the system operating conditions; H W (X) represents the quantitative value of health factors.

[0030] Furthermore, the improved hash function algorithms include the MD5 algorithm, the SHA-1 algorithm, and the HAVAL algorithm.

[0031] Furthermore, the data-driven methods include time series models, hidden Markov models, and logistic regression models.

[0032] The beneficial effects of this application are as follows: The quantitative analysis method for steam turbine units based on hash function dimensionality reduction parameter reconstruction described in this application deeply analyzes the dynamic change characteristics of the entropy of the operating condition fingerprint time series information and the relationship with system health factors, reveals the degradation law of health factors throughout the entire life cycle of the steam turbine unit, the formation mechanism of the increase in the entropy of the operating condition fingerprint of the steam turbine unit, establishes a quantitative assessment model for the health factors of the steam turbine unit, realizes the quantitative calculation of the health factors of the steam turbine unit, solves the application problem of PHM technology in engineering equipment such as steam turbine units, and provides scientific basis and technical support for research in the field of quantitative condition-based maintenance and health management of steam turbine units. It solves the quantitative assessment problem of health management technology in the field of PHM, provides a new idea for the transformation of health management technology for complex systems from qualitative signal analysis to quantitative informatics analysis methods, and provides theoretical reference for the standardization research of PHM technology. Attached Figure Description

[0033] Figure 1 is a flowchart of the quantitative analysis method for steam turbine groups based on hash function-based dimensionality reduction parameter reconstruction as described in this application;

[0034] Figure 2 is a basic flowchart of parameter reconstruction in the embodiments of this application. Detailed Implementation

[0035] The technical solutions of this application will now be described in detail with reference to the accompanying drawings. To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the described embodiments of the present invention are within the scope of protection of the present invention.

[0036] Referring to Figures 1 and 2, an embodiment of this application is provided, which offers a quantitative analysis method for steam turbine units based on hash function-based dimensionality reduction parameter reconstruction. The overall analysis method of this application is shown in Figure 1. First, a study on the classification and usability of multidimensional system operating condition data will be conducted to guide the data collection process. Second, an improved hash function algorithm for dimensionality reduction mapping of multidimensional data will be studied, and a dimensionality-reduced operating condition fingerprint sequence database will be obtained. Subsequently, an analysis of the quantitative calculation of health factors based on the information entropy of the operating condition fingerprint sequence and the correlation analysis method between health level and health factors will be conducted. Finally, a study on a quantitative health factor model for steam turbine units will be carried out, and a prototype testing system will be constructed to conduct online experimental verification.

[0037] As shown in Figure 1, the quantitative analysis method for steam turbine units based on hash function-based dimensionality reduction parameter reconstruction described in this application includes:

[0038] S1: Analyze the types and number of sensors in the experimental setup used for quantitative analysis of steam turbine units to obtain sensor parameters.

[0039] S2: Construct a multidimensional variable storage matrix using historical operating data and sensor parameters.

[0040] S3: An improved hash function algorithm is used to perform collision and perturbation sensitivity analysis on the multidimensional variable storage matrix to obtain one-dimensional working condition fingerprint sequence data.

[0041] Specifically, the improved hash function algorithm includes: normalizing the sampled values ​​of each sensor stored in the multidimensional variable storage matrix to obtain normalization parameters; analyzing the number of significant digits i of the normalization parameters; and then multiplying the normalization parameters by 10i to convert them into large integer values, which are the effective input values ​​of the improved hash function algorithm.

[0042] In this system, the sampled values ​​from each sensor are binary digital signals converted from analog to digital signals (A / D converters). If the analog input of the A / D converter is x, the digital output is y, and the ratio of the transmitter's input range to its output range is K, then... m Then y = x / K m q.

[0043] Where q represents the quantization unit of the A / D conversion, and q = M / 2 NM represents the range of the analog input quantity in the A / D conversion, and N represents the number of bits in the A / D conversion. When converting the analog input quantity to a binary digital quantity, it can only be an integer multiple of q, with the decimal part rounded off. The maximum error of the A / D conversion is ±q / 2. The improved hash function algorithm establishes a one-to-one mapping between the sampled value set and the hash value. Therefore, if a dead zone (also known as a non-operating zone) is not set for the sampled values ​​and the digital quantities within the dead zone are not rounded up or down, the hash value will oscillate repeatedly with changes in the error value. This will cause the fingerprint sequence values ​​of the operating condition information to tend to be evenly distributed, and its information entropy value will tend to be at its maximum. Therefore, it is necessary to conduct in-depth research on the similarity preservation method of sampled values ​​within the dead zone during the design process of the hash function, and introduce a similarity preservation degree variable during data dimensionality reduction to avoid and solve the above problems.

[0044] Based on the above, it can be seen that in the parameter reconstruction method based on hash function projection dimensionality reduction described in this application, the operating conditions of the steam turbine unit are multidimensional data, the dimension of which is determined by the number of sensor measurement points. The steam turbine unit is a complex system, and the variables at each measurement point are not completely independent due to interactive coupling, resulting in significant multicollinearity. Multicollinearity leads to instability in the solution space, resulting in inconsistent results. Parameter reconstruction through dimensionality reduction is an effective method to solve this problem. The characteristic of a hash function is that it transforms an input data sequence of arbitrary length into a hash value of fixed length, mapping data of arbitrary dimensions to one-dimensional data. The hash value generated by compressing multidimensional data of complex system operating conditions can be called a "condition fingerprint." Based on the characteristics of multidimensional data of steam turbine unit operating conditions, this paper will consider factors such as hash collisions, perturbation sensitivity, and the completeness of the condition mapping in parameter reconstruction, and will study in detail an improved hash function suitable for steam turbine unit operating conditions.

[0045] The improved hash function algorithm can store all sensor data under each specific working condition as an array D[m] in the computer, and then map this array D[m] to an integer space represented by 8×n binary bits through a random mapping algorithm; that is, the one-dimensional working condition fingerprint sequence data is to map the array D[m] to an integer space represented by 8×n binary bits.

[0046] The time series fingerprint of operating conditions, composed of integer sequences, can simplify the traditional multidimensional data problem into a one-dimensional variable analysis problem. Since any change in any data in the array D[m] generates a new random large integer, the dimensionality reduction mapping method does not lead to the loss of dynamic information due to local feature parameters for operating condition sequences recorded in chronological order.

[0047] S4: Perform dimensionality reduction projection completeness analysis on the one-dimensional working condition fingerprint sequence data to generate the working condition fingerprint sequence information entropy.

[0048] Specifically, this study investigates a quantitative calculation method for turbine unit health factors based on information entropy. The turbine unit's operating condition fingerprint sequence serves as a storage carrier for the turbine unit's characteristic parameter sequence, recording dynamic information about the system's operating conditions. Information entropy is a quantitative parameter describing system uncertainty; therefore, the information entropy of the turbine unit's operating condition fingerprint sequence can be used as a fundamental parameter for the quantitative calculation of its health factors. Utilizing the large-scale unit structure vibration test system from the National Engineering Center for Thermal Power Unit Vibration, a system degradation model will be established based on historical test data, various experimental data, online condition monitoring, and test data. The quantitative relationship between information entropy and health status will be studied in detail, and the correlation between health factors based on information entropy and the actual health level of the system will be further explored.

[0049] Regarding the representation capability of operating condition information in integer space, if the frequency of a complex system is f, and the system has m sensor measurement points, assuming that the operating conditions collected each time are different, then an n-byte integer space satisfies a non-repeating operating condition mapping for N consecutive years, expressed as:

[0050]

[0051] Assuming f = 10 and m = 50000, even with n = 6, a non-repeating operating condition mapping can theoretically be satisfied for approximately 2495 years. Since current mainstream hash function algorithms support mapping with large integers n = 16, we can conclude that within the finite lifespan of a steam turbine unit, operating condition fingerprints can satisfy both data structure consistency and the completeness of a one-to-one mapping across all operating conditions.

[0052] Since operating condition fingerprints possess data structure consistency and a complete one-to-one mapping across all operating conditions, their sequence necessarily serves as a storage carrier for system-level characteristic parameter sequences, recording information about the dynamic evolution of the macroscopic system's operating state. Therefore, the concept of information entropy is defined, expressed as:

[0053] H(X)=-∑ x∈X P(x)log2P(x);

[0054] Where H(X) represents the information entropy value, and P(x)log2P(x) represents the system state uncertainty.

[0055] S5: Perform quantitative analysis of health factors based on the entropy of the working condition fingerprint sequence information to obtain quantitative values ​​of health factors.

[0056] Specifically, information entropy is a measure of the amount of information required to eliminate system uncertainty. The higher the system's uncertainty, the more information needs to be introduced to eliminate it, resulting in higher information entropy, and vice versa. If system fault diagnosis is not considered, information entropy can be used as a unit of measurement for system state uncertainty; that is, the value of information entropy can also be extended to "system state uncertainty." Therefore, if a relationship model can be established between "system state uncertainty" and the system's operational health, then "information entropy" can be used as a fundamental parameter to describe the system's health, i.e., its health factor can be quantitatively calculated.

[0057] However, in practical applications, the normal operating conditions of complex industrial systems far outnumber the fault conditions. According to the calculation formula of information entropy, if the information entropy formula is applied directly, then H≈0. Information entropy is not sensitive enough to changes in the state of the system, which will directly affect the accuracy of health factor calculation.

[0058] This application emphasizes the use of data-driven methods such as time series models, hidden Markov models, and logistic regression models to dynamically weight the statistical probabilities in the information entropy calculation formula, as shown below:

[0059] H W (X)=-∑ x∈X W(P)P(x)log2[W(P)P(x)];

[0060] Where W(P) represents the probability statistical distribution function of the system operating conditions; H W (X) represents the quantitative value of health factors.

[0061] Using the dynamic value sequence of the aforementioned health factors as the data source, online experiments were conducted to analyze and calculate the effectiveness of the quantitative calculation method for health factors and the system performance degradation model.

[0062] S6: Analyze the correlation between quantitative values ​​of health factors and health status based on data-driven methods to construct a quantitative health factor model for steam turbine units.

[0063] S7: Conduct online verification experiments on the quantitative health factor model of the turbine group using historical operating data to optimize and improve the quantitative health factor model of the turbine group.

[0064] S8: Quantitative analysis of the turbine set is conducted using the optimized and improved quantitative health factor model.

[0065] Based on the above technical solutions, the core of health management in this application lies in the monitoring and analysis of system degradation performance, and does not necessarily require the injection of fault data. Furthermore, the number of characteristic variables in turbine operating condition parameters, which possess multidimensional data characteristics, can be kept within a certain range through dimensionality reduction methods. Therefore, data-driven health management is one of the feasible solutions to the problem of quantitative calculation of health factors in turbine units. Dimensionality reduction preprocessing of high-dimensional measurement point variables can effectively simplify the problem of quantitative calculation of health factors.

[0066] In summary, this application, through theoretical deduction and experimental verification, deeply analyzes the dynamic changes in the entropy of the operating condition fingerprint time series information and the relationship between system health factors. It reveals the degradation law of health factors throughout the entire life cycle of steam turbine units, the formation mechanism of the increase in the entropy of the operating condition fingerprint complexity of steam turbine units, establishes a quantitative health factor model for steam turbine units, realizes the quantitative calculation of steam turbine unit health factors, and solves the application problem of PHM technology in engineering equipment such as steam turbine units. This provides a scientific basis and technical support for research in the field of quantitative condition-based maintenance and health management of steam turbine units. In conclusion, this invention solves the quantitative assessment problem of health management technology in the field of PHM, provides a new approach for the transformation of health management technology for complex systems from qualitative signal analysis to quantitative informatics analysis methods, and provides a theoretical reference for the standardization research of PHM technology.

[0067] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A quantitative analysis method for steam turbine units based on hash function-based dimensionality reduction parameter reconstruction, characterized in that, include: S1: Analyze the types and number of sensors in the experimental setup used for quantitative analysis of steam turbine units to obtain sensor parameters; S2: Construct a multidimensional variable storage matrix using historical operating data and sensor parameters; S3: Perform collision and perturbation sensitivity analysis on the multidimensional variable storage matrix using an improved hash function algorithm to obtain one-dimensional operating condition fingerprint sequence data; S4: Perform dimensionality reduction projection completeness analysis on the one-dimensional operating condition fingerprint sequence data to generate operating condition fingerprint sequence information entropy. S5: Perform quantitative analysis of health factors based on the entropy of the working condition fingerprint sequence information to obtain quantitative values ​​of health factors; S6: Analyze the correlation between quantitative values ​​of health factors and health status based on data-driven methods to construct a quantitative health factor model for steam turbine units; S7: Conduct online verification experiments on the quantitative health factor model of the turbine set using historical operating data to optimize and improve the quantitative health factor model of the turbine set; S8: Quantitative analysis of the turbine group is performed using the optimized and improved quantitative health factor model; wherein, in step S3, the improved hash function algorithm includes: normalizing the sampled values ​​of each sensor stored in the multidimensional variable storage matrix to obtain normalization parameters; analyzing the effective number of digits i of the normalization parameters, and then multiplying the normalization parameters by 10i to convert them into large integer values, which are the effective input values ​​of the improved hash function algorithm; wherein, the sampled values ​​of each sensor are all binary digital signals after A / D conversion, if the analog input of the A / D conversion is... The output digital quantity is The ratio of the transmitter's input range to its output range is ,but ;in, This represents the quantization unit of the A / D conversion, and ， This indicates the range of the analog input quantity in the A / D conversion. Indicates the number of bits used in the A / D conversion.

2. The method as described in claim 1, characterized in that, In step S3, the one-dimensional working condition fingerprint sequence data is obtained by dividing the array... Mapped to a by A space of integers represented by binary bits.

3. The method as described in claim 1, characterized in that, In step S4, the entropy of the working condition fingerprint sequence information is represented as: (1) Among them, Represents the information entropy value. It represents the uncertainty of the system state.

4. The method as described in claim 3, characterized in that, In step S5, the quantitative value of the health factor is obtained by dynamically weighting the statistical probability in equation (1), and is expressed as: ;in, The probability and statistical distribution function representing the system's operating conditions; This represents the quantitative value of health factors.

5. The method as described in claim 1, characterized in that, The improved hash function algorithms include MD5, SHA-1, and HAVAL.

6. The method as described in claim 1, characterized in that, The data-driven methods include time series models, hidden Markov models, and logistic regression models.

Images